[PATCHv6 2/7] crypto: sa2ul: Add crypto driver

[Nathan Chancellor]

On Mon, Jul 13, 2020 at 11:34:22AM +0300, Tero Kristo wrote:
> From: Keerthy <j-keerthy at ti.com>
> 
> Adds a basic crypto driver and currently supports AES/3DES
> in cbc mode for both encryption and decryption.
> 
> Signed-off-by: Keerthy <j-keerthy at ti.com>
> [t-kristo at ti.com: major re-work to fix various bugs in the driver and to
>  cleanup the code]
> Signed-off-by: Tero Kristo <t-kristo at ti.com>
> ---
>  drivers/crypto/Kconfig  |   14 +
>  drivers/crypto/Makefile |    1 +
>  drivers/crypto/sa2ul.c  | 1388 +++++++++++++++++++++++++++++++++++++++
>  drivers/crypto/sa2ul.h  |  380 +++++++++++
>  4 files changed, 1783 insertions(+)
>  create mode 100644 drivers/crypto/sa2ul.c
>  create mode 100644 drivers/crypto/sa2ul.h

<snip>

> diff --git a/drivers/crypto/sa2ul.c b/drivers/crypto/sa2ul.c
> new file mode 100644
> index 000000000000..860c7435fefa
> --- /dev/null
> +++ b/drivers/crypto/sa2ul.c
> @@ -0,0 +1,1388 @@

<snip>

> +static int sa_run(struct sa_req *req)
> +{
> +	struct sa_rx_data *rxd;
> +	gfp_t gfp_flags;
> +	u32 cmdl[SA_MAX_CMDL_WORDS];
> +	struct sa_crypto_data *pdata = dev_get_drvdata(sa_k3_dev);
> +	struct device *ddev;
> +	struct dma_chan *dma_rx;
> +	int sg_nents, src_nents, dst_nents;
> +	int mapped_src_nents, mapped_dst_nents;
> +	struct scatterlist *src, *dst;
> +	size_t pl, ml, split_size;
> +	struct sa_ctx_info *sa_ctx = req->enc ? &req->ctx->enc : &req->ctx->dec;
> +	int ret;
> +	struct dma_async_tx_descriptor *tx_out;
> +	u32 *mdptr;
> +	bool diff_dst;
> +	enum dma_data_direction dir_src;
> +
> +	gfp_flags = req->base->flags & CRYPTO_TFM_REQ_MAY_SLEEP ?
> +		GFP_KERNEL : GFP_ATOMIC;
> +
> +	rxd = kzalloc(sizeof(*rxd), gfp_flags);
> +	if (!rxd)
> +		return -ENOMEM;
> +
> +	if (req->src != req->dst) {
> +		diff_dst = true;
> +		dir_src = DMA_TO_DEVICE;
> +	} else {
> +		diff_dst = false;
> +		dir_src = DMA_BIDIRECTIONAL;
> +	}
> +
> +	/*
> +	 * SA2UL has an interesting feature where the receive DMA channel
> +	 * is selected based on the data passed to the engine. Within the
> +	 * transition range, there is also a space where it is impossible
> +	 * to determine where the data will end up, and this should be
> +	 * avoided. This will be handled by the SW fallback mechanism by
> +	 * the individual algorithm implementations.
> +	 */
> +	if (req->size >= 256)
> +		dma_rx = pdata->dma_rx2;
> +	else
> +		dma_rx = pdata->dma_rx1;
> +
> +	ddev = dma_rx->device->dev;
> +
> +	memcpy(cmdl, sa_ctx->cmdl, sa_ctx->cmdl_size);
> +
> +	sa_update_cmdl(req, cmdl, &sa_ctx->cmdl_upd_info);
> +
> +	if (req->type != CRYPTO_ALG_TYPE_AHASH) {
> +		if (req->enc)
> +			req->type |=
> +				(SA_REQ_SUBTYPE_ENC << SA_REQ_SUBTYPE_SHIFT);
> +		else
> +			req->type |=
> +				(SA_REQ_SUBTYPE_DEC << SA_REQ_SUBTYPE_SHIFT);
> +	}
> +
> +	cmdl[sa_ctx->cmdl_size / sizeof(u32)] = req->type;
> +
> +	/*
> +	 * Map the packets, first we check if the data fits into a single
> +	 * sg entry and use that if possible. If it does not fit, we check
> +	 * if we need to do sg_split to align the scatterlist data on the
> +	 * actual data size being processed by the crypto engine.
> +	 */
> +	src = req->src;
> +	sg_nents = sg_nents_for_len(src, req->size);
> +
> +	split_size = req->size;
> +
> +	if (sg_nents == 1 && split_size <= req->src->length) {
> +		src = &rxd->rx_sg;
> +		sg_init_table(src, 1);
> +		sg_set_page(src, sg_page(req->src), split_size,
> +			    req->src->offset);
> +		src_nents = 1;
> +		dma_map_sg(ddev, src, sg_nents, dir_src);
> +	} else {
> +		mapped_src_nents = dma_map_sg(ddev, req->src, sg_nents,
> +					      dir_src);
> +		ret = sg_split(req->src, mapped_src_nents, 0, 1, &split_size,
> +			       &src, &src_nents, gfp_flags);
> +		if (ret) {
> +			src_nents = sg_nents;
> +			src = req->src;
> +		} else {
> +			rxd->split_src_sg = src;
> +		}
> +	}
> +
> +	if (!diff_dst) {
> +		dst_nents = src_nents;
> +		dst = src;
> +	} else {
> +		dst_nents = sg_nents_for_len(req->dst, req->size);
> +
> +		if (dst_nents == 1 && split_size <= req->dst->length) {
> +			dst = &rxd->tx_sg;
> +			sg_init_table(dst, 1);
> +			sg_set_page(dst, sg_page(req->dst), split_size,
> +				    req->dst->offset);
> +			dst_nents = 1;
> +			dma_map_sg(ddev, dst, dst_nents, DMA_FROM_DEVICE);
> +		} else {
> +			mapped_dst_nents = dma_map_sg(ddev, req->dst, dst_nents,
> +						      DMA_FROM_DEVICE);
> +			ret = sg_split(req->dst, mapped_dst_nents, 0, 1,
> +				       &split_size, &dst, &dst_nents,
> +				       gfp_flags);
> +			if (ret) {
> +				dst_nents = dst_nents;

This causes a clang warning:

drivers/crypto/sa2ul.c:1152:15: warning: explicitly assigning value of
variable of type 'int' to itself [-Wself-assign]
                                dst_nents = dst_nents;
                                ~~~~~~~~~ ^ ~~~~~~~~~
1 warning generated.

Was the right side supposed to be something else? Otherwise, this line
can be removed, right?

> +				dst = req->dst;
> +			} else {
> +				rxd->split_dst_sg = dst;
> +			}
> +		}
> +	}
> +
> +	if (unlikely(src_nents != sg_nents)) {
> +		dev_warn_ratelimited(sa_k3_dev, "failed to map tx pkt\n");
> +		ret = -EIO;
> +		goto err_cleanup;
> +	}
> +
> +	rxd->tx_in = dmaengine_prep_slave_sg(dma_rx, dst, dst_nents,
> +					     DMA_DEV_TO_MEM,
> +					     DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
> +	if (!rxd->tx_in) {
> +		dev_err(pdata->dev, "IN prep_slave_sg() failed\n");
> +		ret = -EINVAL;
> +		goto err_cleanup;
> +	}
> +
> +	rxd->req = (void *)req->base;
> +	rxd->enc = req->enc;
> +	rxd->ddev = ddev;
> +	rxd->src = src;
> +	rxd->dst = dst;
> +	rxd->iv_idx = req->ctx->iv_idx;
> +	rxd->enc_iv_size = sa_ctx->cmdl_upd_info.enc_iv.size;
> +	rxd->tx_in->callback = req->callback;
> +	rxd->tx_in->callback_param = rxd;
> +
> +	tx_out = dmaengine_prep_slave_sg(pdata->dma_tx, src,
> +					 src_nents, DMA_MEM_TO_DEV,
> +					 DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
> +
> +	if (!tx_out) {
> +		dev_err(pdata->dev, "OUT prep_slave_sg() failed\n");
> +		ret = -EINVAL;
> +		goto err_cleanup;
> +	}
> +
> +	/*
> +	 * Prepare metadata for DMA engine. This essentially describes the
> +	 * crypto algorithm to be used, data sizes, different keys etc.
> +	 */
> +	mdptr = (u32 *)dmaengine_desc_get_metadata_ptr(tx_out, &pl, &ml);
> +
> +	sa_prepare_tx_desc(mdptr, (sa_ctx->cmdl_size + (SA_PSDATA_CTX_WORDS *
> +				   sizeof(u32))), cmdl, sizeof(sa_ctx->epib),
> +			   sa_ctx->epib);
> +
> +	ml = sa_ctx->cmdl_size + (SA_PSDATA_CTX_WORDS * sizeof(u32));
> +	dmaengine_desc_set_metadata_len(tx_out, req->mdata_size);
> +
> +	dmaengine_submit(tx_out);
> +	dmaengine_submit(rxd->tx_in);
> +
> +	dma_async_issue_pending(dma_rx);
> +	dma_async_issue_pending(pdata->dma_tx);
> +
> +	return -EINPROGRESS;
> +
> +err_cleanup:
> +	dma_unmap_sg(ddev, req->src, sg_nents, DMA_TO_DEVICE);
> +	kfree(rxd->split_src_sg);
> +
> +	if (req->src != req->dst) {
> +		dst_nents = sg_nents_for_len(req->dst, req->size);
> +		dma_unmap_sg(ddev, req->dst, dst_nents, DMA_FROM_DEVICE);
> +		kfree(rxd->split_dst_sg);
> +	}
> +
> +	kfree(rxd);
> +
> +	return ret;
> +}

Cheers,
Nathan